An RF plasma reactor is used to process semiconductor wafers to produce microelectronic circuits. The reactor forms a plasma within a chamber containing the wafer to be processed. The plasma is formed and maintained by application of RF plasma source power coupled either inductively or capacitively into the chamber. For capacitive coupling of RF source power into the chamber, an overhead electrode (facing the wafer) is powered by an RF source power generator.
One problem in such reactors is that the output impedance of the RF generator, typically 50 Ohms, must be matched to the load impedance presented by the combination of the electrode and the plasma. Otherwise the amount of RF power delivered to the plasma chamber will fluctuate with fluctuations in the plasma load impedance so that certain process parameters such as plasma density cannot be held within the required limits. The plasma load impedance fluctuates during processing because it depends upon conditions inside the reactor chamber which tend to change dynamically as processing progresses. At an optimum plasma density for dielectric or metal etch processes, the load impedance is very small compared to the output impedance of the RF generator and can vary significantly during the processing of the wafer. Accordingly, an impedance match circuit must be employed to actively maintain an impedance match between the generator and the load. Such active impedance matching uses either a variable reactance and/or a variable frequency. One problem with such impedance match circuits is that they must be sufficiently agile to follow rapid changes in the plasma load impedance, and therefore are relatively expensive and can reduce system reliability due to their complexity.
Another problem is that the range of load impedances over which the match circuit can provide an impedance match (the “match space”) is limited. The match space is related to the system Q, where Q=f/f, f being a resonant frequency of the system and f being the bandwidth on either side of f within which resonant amplitude is within 6 dB of the peak resonant amplitude at f. The typical RF generator has a limited ability to maintain the forward power at a nearly constant level even as more RF power is reflected back to the generator as the plasma impedance fluctuates. Typically, this is achieved by the generator servoing its forward power level, so that as an impedance mismatch increases (and therefore reflected power increases), the generator increases its forward power level. Of course, this ability is limited by the maximum forward power [of] which the generator is capable of producing. Typically, the generator is capable of handling a maximum ratio of forward standing wave voltage to reflected wave voltage (i.e., the voltage standing wave ratio or VSWR) of not more than 3:1. If the difference in impedances increases (e.g., due to plasma impedance fluctuations during processing) so that the VSWR exceeds 3:1, then the RF generator can no longer control the delivered power, and control over the plasma is lost. As a result, the process is likely to fail. Therefore, at least an approximate impedance match must be maintained between the RF generator and the load presented to it by the combination of the coil antenna and the chamber. This approximate impedance match must be sufficient to keep the VSWR at the generator output within the 3:1 VSWR limit over the entire anticipated range of plasma impedance fluctuations. The impedance match space is, typically, the range of load impedances for which the match circuit can maintain the VSWR at the generator output at or below 3:1.
A related problem is that the load impedance itself is highly sensitive to process parameters such as chamber pressure, plasma source power level, source power frequency and plasma density. This limits the range of such process parameters (the “process window”) within which the plasma reactor must be operated to avoid an unacceptable impedance mismatch or avoid fluctuations that take load impedance outside of the match space. Likewise, it is difficult to provide a reactor which can be operated outside of a relatively narrow process window and process use, or one that can handle many process applications.
Another related problem is that the load impedance is also affected by the configuration of the reactor itself, such as dimensions of certain mechanical features and the conductivity or dielectric constant of certain materials within the reactor. (Such configurational items affect reactor electrical characteristics, such as stray capacitance for example, that in turn affect the load impedance.) This makes it difficult to maintain uniformity among different reactors of the same design due to manufacturing tolerances and variations in materials. As a result, with a high system Q and correspondingly small impedance match space, it is difficult to produce any two reactors of the same design which exhibit the same process window or provide the same performance.
Another problem is inefficient use of the RF power source. Plasma reactors are known to be inefficient, in that the amount of power delivered to the plasma tends to be significantly less than the power produced by the RF generator. As a result, an additional cost in generator capability and a trade-off against reliability must be incurred to produce power in excess of what is actually required to be delivered into the plasma.
This specification generally concerns a capacitively coupled reactor having an overhead electrode that is driven by a VHF plasma source power RF supply, rather than a HF power supply. We have found that at a VHF source power frequency, unlike an HF frequency, it is practical to tune the overhead electrode to resonate with the plasma, leading to heretofore unattained stability and efficiency and many other advantages.
A superior way to feed process gases into a capacitively coupled plasma reactor is to inject the process gases through the overhead ceiling. In the present reactor, the overhead ceiling is the source power electrode that is coupled to a VHF RF power generator through a coaxial stub or equivalent impedance match element. In order to inject the process gas from the ceiling, the ceiling electrode also is a “showerhead”, a conductive layer having a set of small gas injection ports passing through it for injecting the process gases. In some instances plasma discharge or “arcing” tends to occur within the ceiling gas injection ports. This poses a risk of the plasma discharge removing material of the overhead electrode or showerhead from the interior surfaces of the gas injection ports. The species, e.g. metallic species, thus introduced into the plasma can contaminate the surface of the wafer or workpiece and damage the microelectronic devices being fabricated thereon.
Accordingly, it would be desirable to reduce or eliminate the tendency to ignite plasma within gas injection ports (or anywhere else outside of the bulk plasma), particularly in a plasma reactor having a combination overhead electrode/gas distribution showerhead connected to a VHF plasma source RF power supply.
The overhead electrode may suffer wear from being in contact with plasma, particularly since it is both an anode for the bias power and a cathode for the source power and is therefore subject to RF and DC currents. The cost of operating the reactor would be reduced if there could be found a way to avoid passing such currents directly through the overhead electrode or avoid direct contact of plasma with the electrode.
A problem often found in plasma reactors is the generation of second and third harmonics within the plasma sheath. In the present reactor, while plasma VHF source power is applied by the overhead electrode, plasma bias power is applied by an HF signal on the wafer support pedestal. At HF frequencies, most of the RF power is consumed in the sheath, the remainder sustaining the bulk plasma. A plasma sheath is a non-linear load and therefore creates second and/or third harmonics of the HF bias signal applied to the wafer support pedestal. The presence of such harmonics changes plasma behavior in such a way that process performance is impaired in the presence of such harmonics. Specifically, process control to avoid etch stop or over-etching becomes more difficult, and the etch rate is reduced. It would be desirable to reduce the production of such harmonics affecting the plasma.
In the present reactor, the coaxial tuning stub, whose length is correlated to the wavelength of the VHF source power signal, can have a footprint larger than the remainder of the reactor. It would therefore be advantageous to be able to reduce this footprint without sacrificing any of the advantages of the coaxial tuning stub.